Attention-Based Deep Multi-Instance Learning for Weakly-labeled Breast Ultrasound Image Classification

This project is part of a larger initiative aimed at improving breast cancer diagnosis. The specific focus of this project is to develop and implement a deep multi-instance learning (DML) model that can improve the accuracy of breast cancer diagnosis in ultrasound images. The objectives of this project are: 

– To address challenges presented by weakly labeled medical images. In medical images, conventional machine-learning approaches encounter limitations because fully annotated datasets are not always available. 

 - To add interpretability to the DML model. Attention-based techniques are incorporated in the proposed DML model to ensure transparency in healthcare professionals’ decision-making processes.